AI is becoming the standard for financial investigations, but uptake varies amongst other industries. We interviewed accredited fraud investigator Tim McBride of Mastercard and asked him to explain how AI supercharges financial investigations.
Financial investigators have to be flexible, multi-skilled, intuitive and decisive. They follow leads, write computer code and analyze a lot of data. A lot of data! But they can’t rely on instinct and rules-based decisioning alone. That AI supercharges financial investigations is a given, says veteran investigator Tim McBride. It’s a vital tool that saves – even makes – organizations money, saves time and finds fraud that humans may not recognize. Here are Tim’s top 5 ways AI to supercharge your team.
#1 AI reduces the need for rules
Rules-based systems take constant care and feeding, with investigators inputting each finding, newly identified fraud schemes, and new parameters for identifying fraud and risk. True, rules are easy to code, but when a system has thousands of lines of code supporting 50,000 rules, how does one know what’s current? And if one rule is deleted, it can have unforeseen consequences by affecting other rules that depend on it. Including historical data.
AI models are built using a few rules plus algorithms that are coded to identify more complex fraud. It uses labeled data and teaches the model all the rules attached to that data. The data need occasional updates, but the beauty of AI is that it continually updates independent of humans. It does need monitoring, of course, and if rules or regulations change, those need to be manually updated and old code removed.
In a rules-based system, filtering and decisioning millions of transactions takes significant time. With AI, your transaction is filtered in nanoseconds and gives real-time results to approve or halt the transaction at source.
Benefit: Less maintenance (Our technology reduced 50,000 rules to 250 for Worldpay, the world’s largest acquirer.)
#2 AI enables continuous learning
Feedback is a critical piece of keeping fraud and risk detection solutions relevant. Without feedback and updates, your solution quickly becomes dated. AI models record the nuances of change with businesses and their customers through continual updates. New trends and patterns are recognized and saved in the model, identifying “good” and “bad” behaviors.
An AI model improves with age as it learns more about your business, your expected outcomes and what normal transaction behavior looks like. Think of a continual feedback loop:
- Input data: Raw data is imported in any format, then enriched by AI tools (algorithms).
- Model development: Data is scored to create a baseline.
- Enhanced pattern recognition: Supervised and unsupervised learning reconcile the outputs to meet the goals; may be intercepted by an investigator.
- Incremental learning: Continuously learns from new transactions and refines the model.
- Repeat: The cycle continues, refining the model in real time to increase efficiency.
Benefit: Accurate results
#3 AI reduces false positives
Consider this: 44 percent of consumers stopped shopping at a retailer after receiving a false decline. Tim McBride estimates he spent 40 percent of his time as an investigator chasing down false positives and leads on fraud. Doing the same work with machine learning reduced that time to 20 percent. He focused his found time on investigating complex fraud schemes, often worth hundreds of thousands of dollars. His entire unit transformed into a strong revenue generator.
The number of false positives is influenced by how the model is built. It should use a variety of tools to solve your organization’s unique business problem, then be tested for results, making adjustments throughout development.
Brighterion AI reduced Worldpay’s false positives by 20x, a huge gain in revenue for its merchants.
Benefit: Increased efficiencies
#4 AI allows real-time detection
Continuous updates and reduced false positives enable real-time detection. Regular model updates and pattern recognition enables fraudulent transactions to be denied before they are processed. In industries such as healthcare, where investigations traditionally take place after claims are paid, this totally transforms the industry. In direct payment transactions, fraud and chargebacks are avoided as merchant accounts are flagged in real time.
An important aspect of real-time detection is to use a solution that is supported by a distributed file system. There are no single points of failure and users can expect 99.9999 percent uptime. This also enables extraordinary scalability as no one server is being overburdened.
Benefit: 24/7 protection
#5 AI uses one-to-one analysis
The biggest advantage of AI is that it identifies patterns that humans don’t recognize. While people define the problem to be solved and thresholds for the model, humans are limited by our own knowledge, experience and biases. (Thresholds refers to the parameters for investigative minimums, enabling investigators to focus on substantial cases.)
AI’s algorithms identify and compare every data point within a transaction, looking for behaviors and patterns. They compare frequency, transaction size, location, health diagnoses, sudden changes in credit or payment behavior and more. The data is then compared to results from other merchants or service providers to identify anomalous behavior. AI reduces hours of investigation to milliseconds.
Benefit: Wholistic detection
Visit the Brighterion industry pages to learn more about how AI supercharges financial investigations in your sector.
Tim McBride, AHFI, PMC3 is an accredited investigator with the National Health Care Anti-Fraud Association. He is the Director, Product Development, Cyber and Intelligence Solutions, Mastercard.